CN115963407A - A lithium battery SOC estimation method based on ICGWO optimized ELM - Google Patents

Abstract

The present invention provides a lithium battery SOC estimation method based on ICGWO optimized ELM, which includes the following steps: Step 1: Aiming at the problems that the gray wolf algorithm often encounters premature phenomenon and local convergence due to the rapid decline of population diversity in the later stage of evolution, through Introduce an improved strategy to construct an improved gray wolf algorithm with good global optimization performance; step 2: use the improved gray wolf algorithm, and take the minimum root mean square error of the model output as the objective function, and set the threshold value of the hidden layer of the extreme learning machine and The input weight parameters are optimized, and the lithium battery SOC estimation model based on ICGWO optimized ELM is established; the application of this technical solution can achieve better prediction accuracy and generalization ability, and can provide important feedback information for the battery management system of new energy vehicles.

Description

A lithium battery SOC estimation method based on ICGWO optimized ELM

Technical Field

The invention relates to the technical field of battery system management, and in particular to a lithium battery SOC estimation method based on ICGWO optimized ELM.

Background Art

Pure electric vehicles (EV) and hybrid electric vehicles (HEV) are becoming more and more popular as an alternative means of transportation. They are environmentally friendly, have no exhaust pollution, smooth speed switching, and small vibration amplitude. The main problem currently faced by EV and HEV is that the overall cruising range is significantly shorter than that of traditional ICE vehicles. The lack of a battery management system that can effectively estimate and predict the actual remaining power of the battery is the bottleneck for increasing the cruising range. In order to describe the state of lithium-ion batteries storing electrical energy, an indicator to measure the remaining power of the battery is proposed to represent the battery state of charge (SOC). Klein et al. gave the definition of battery SOC: the ratio of the available battery capacity to the nominal battery capacity at a given time, 100% means that the battery is fully charged, and 0% means that it is empty. How to accurately determine and calculate SOC has become the key to building a battery management system (BMS).

In recent years, with the rapid development of artificial intelligence and machine learning theory, data-driven intelligent modeling methods represented by neural networks and support vector machines have become a mainstream method for battery SOC estimation. Extreme Learning Machine (ELM) is a new single hidden layer feedforward neural network model (SLFNs) with excellent performance proposed based on generalized inverse matrix theory. Compared with traditional neural networks and support vector machine regression models, ELM has the advantages of simple mathematical model, fast learning speed and strong generalization ability. It has been widely used in pattern recognition, fault diagnosis and other fields, and has become an important research direction in the field of lithium battery SOC estimation. However, the inventors found that the charging and discharging process of lithium batteries has the characteristics of complex mechanism and strong nonlinearity. Many factors will affect the charging and discharging efficiency, and then affect the accuracy of the SOC estimation model. Traditional extreme learning machines may fall into local optimal solutions because the loss function is not a convex function. At the same time, if the algorithm parameters are not set properly, it may cause overfitting, which is a problem of poor generalization ability.

Summary of the invention

In view of this, the purpose of the present invention is to provide a lithium battery SOC estimation method based on ICGWO optimized ELM, to achieve better prediction accuracy and generalization ability, and to provide important feedback information for the battery management system of new energy vehicles.

To achieve the above object, the present invention adopts the following technical solution: a lithium battery SOC estimation method based on ICGWO optimized ELM, comprising the following steps:

Step 1: In view of the fact that the gray wolf algorithm often encounters premature phenomenon and local convergence problems due to the rapid decline of population diversity in the late stage of evolution, an improved gray wolf algorithm with good global optimization performance is constructed by introducing an improvement strategy;

Step 2: Using the improved grey wolf algorithm, with the minimization of the root mean square error of the model output as the objective function, the hidden layer threshold and input weight parameters of the extreme learning machine are optimized to establish a lithium battery SOC estimation model based on ICGWO optimized ELM.

In a preferred embodiment: in step 1, the improvement strategy of the ICGWO algorithm is:

Using the randomness and ergodicity of chaotic sequences, the Tent mapping equation is selected to generate the initial population; the algorithm parameters are set and the optimal first three positions α, β and δ under the current number of iterations are recorded, and the rest are recorded as ω wolves;

The location of the wolf pack is determined according to the nonlinear factor adjustment strategy to increase the local search ability of the algorithm; finally, the current optimal solution is perturbed by Cauchy mutation to increase population diversity and improve the global search ability of the algorithm.

In a preferred embodiment: in step 1, the specific steps of the ICGWO algorithm are:

Step S11: setting the initial population size N, the maximum number of iterations T _max , the adjustment coefficient λ and other parameters of the ICGWO algorithm, and determining the upper and lower limits of the optimization variables;

Step S12: using the Tent chaotic mapping formula to generate an initial gray wolf population that satisfies the upper and lower limit constraints of the variables, and setting t=1;

Where _xt is the position of the tth wolf;

Step S13: Calculate the fitness of each wolf species, select the first three wolves with the lowest fitness and record them as α, β and δ respectively, and the rest are recorded as ω wolves;

Step S14: Determine whether the maximum number of iterations has been reached. If yes, output the fitness corresponding to the α wolf and the algorithm ends. Otherwise, jump to step S15.

计算非线性收敛因子a，然后根据式A＝2a·r₁-a和C＝2·r₂计算出系数向量A和C；Step S15: According to the formula

Calculate the nonlinear convergence factor a, and then calculate the coefficient vectors A and C according to the formula A = 2a·r ₁ -a and C = 2·r ₂ ;

Where a is the nonlinear convergence factor, A and C are coefficient vectors;

Step S16: Update the positions of the various species of wolves in the gray wolf pack in sequence according to the following formula:

D＝|C· _Xp (t)-X(t)|

X(t+1)＝ _Xp (t)-A·D

Where t is the current iteration number, D is the distance between the wolf and the prey, A and C are coefficient vectors, _Xp (t) is the position of the prey, and X is the position of the wolf. In particular, the starting point of the wolf pack in the next iteration is determined by the following formula:

Where X ₁ , X ₂ , and X ₃ are the α, β, and δ wolf positions, respectively;

Step S17: Perform Cauchy mutation operation on α wolf according to the following formula to determine whether the maximum number of iterations has been reached. If not, return to step S12 to continue the calculation;

_Xg (t)＝ _Xg (t)+η×C(0,1)

in

Where _Xg (t) is the global optimal solution in the tth generation; η is the mutation weight; λ is the adjustment factor; C(0,1 is the random number of the Cauchy mutation with the proportional parameter t=1.

In a preferred embodiment: in step 2, the objective function of the ICGWO algorithm is:

The mean square error between the ELM model predicted output value and the measured value reflects the model prediction accuracy and generalization performance. Therefore, this paper selects the optimization objective function to minimize the ELM output root mean square error f _RMSE , which can effectively improve the fitting accuracy and generalization performance of the ELM regression model; the objective function is defined as follows:

Wherein, j=1,…,N, L is the number of hidden layer nodes, g(x) is the activation function, _wj =[ _wj1 , _wj2 ,…, _wjn ] ^T is the weight between the jth hidden layer node and the input node, _βj =[ _βj1 , _βj2 ,…, _βjm ^{] T} is the weight between the jth hidden layer node and the output node, _bj is the threshold of the jth hidden layer node; _xi =[ _xi1 , _xi2 ,…, _xin ] ^T∈Rn represents the input data, _yi = _yi1 , _yi2 ,…, _yim ^{] T∈Rm} represents the expected output value.

Compared with the prior art, the present invention has the following beneficial effects:

(1) The present invention combines the improved grey wolf algorithm with ELM and utilizes the global search capability of the improved grey wolf algorithm to optimize the parameters of the ELM, thereby avoiding the problem that the ELM algorithm is prone to falling into local optimality.

(2) The present invention uses Tent chaotic mapping to generate the initial population of the gray wolf algorithm, so that the initial individuals will be distributed as evenly as possible in the search area, thereby improving the diversity and adaptability of the initial population and accelerating the population evolution process.

(3) The present invention adopts a nonlinear adjustment strategy to design the convergence factor, so that the convergence factor decreases slowly in the early stage, which is conducive to enhancing the global exploration ability of the algorithm; and the decrease speed is accelerated in the later stage, which can effectively improve the convergence of the algorithm.

(4) The present invention introduces the Cauchy mutation operator to ensure the diversity of the Grey Wolf Algorithm population and avoid the occurrence of premature phenomenon.

(5) The present invention has better convergence accuracy and faster convergence speed. The results show that the SOC prediction method based on the ICGWO-ELM model of the present invention has better prediction accuracy and generalization ability, and can provide important feedback information for the battery management system of new energy vehicles.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of an algorithm according to a preferred embodiment of the present invention.

FIG. 2 is an ICGWO-ELM model process of a preferred embodiment of the present invention.

FIG. 3 is a time-speed diagram of a simulated driving condition according to a preferred embodiment of the present invention.

FIG. 4 is a diagram comparing training results of different algorithms according to a preferred embodiment of the present invention.

FIG. 5 is a comparison diagram of estimation results of different algorithms according to a preferred embodiment of the present invention.

FIG. 6 is a diagram comparing relative errors of different algorithms according to a preferred embodiment of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and/or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and/or their combinations.

A lithium battery SOC estimation method based on ICGWO optimized ELM, referring to Figures 1 to 6, comprises the following steps:

Step 1: In view of the fact that the gray wolf algorithm often encounters premature phenomenon and local convergence problems due to the rapid decline of population diversity in the late stage of evolution, an improved gray wolf algorithm with good global optimization performance is constructed by introducing an improvement strategy;

Step 2: Using the improved grey wolf algorithm, with the minimization of the root mean square error of the model output as the objective function, the hidden layer threshold and input weight parameters of the extreme learning machine are optimized to establish a lithium battery SOC estimation model based on ICGWO optimized ELM.

In step 1, the main improvement strategy of the ICGWO algorithm is:

Using the randomness and ergodicity of chaotic sequences, the Tent mapping equation is selected to generate the initial population.

Secondly, the wolf pack location is determined according to the nonlinear factor adjustment strategy to increase the local search capability of the algorithm. Finally, the current optimal solution is perturbed by Cauchy mutation to increase population diversity and improve the global search capability of the algorithm.

In step 1, the specific steps of the ICGWO algorithm are:

Step S11: setting the initial population size N, the maximum number of iterations T _max , the adjustment coefficient λ and other parameters of the ICGWO algorithm, and determining the upper and lower limits of the optimization variables;

Step S12: using the Tent chaotic mapping formula to generate an initial gray wolf population that satisfies the upper and lower limit constraints of the variables, and setting t=1;

Where _xt is the position of the tth wolf;

Step S13: Calculate the fitness of each wolf species, select the first three wolves with the lowest fitness and record them as α, β and δ respectively, and the rest are recorded as ω wolves;

Step S14: Determine whether the maximum number of iterations has been reached. If yes, output the fitness corresponding to the α wolf and the algorithm ends. Otherwise, jump to step S15.

计算非线性收敛因子a，然后根据式A＝2a·r₁-a和C＝2·r₂计算出系数向量A和C；Step S15: According to the formula

Calculate the nonlinear convergence factor a, and then calculate the coefficient vectors A and C according to the formula A = 2a·r ₁ -a and C = 2·r ₂ ;

Where a is the nonlinear convergence factor, A and C are coefficient vectors;

Step S16: Update the positions of the various species of wolves in the gray wolf pack in sequence according to the following formula:

D＝|C· _Xp (t)-X(t)|

X(t+1)＝ _Xp (t)-A·D

Where t is the current iteration number, D is the distance between the wolf and the prey, A and C are coefficient vectors, _Xp (t) is the position of the prey, and X is the position of the wolf. In particular, the starting point of the wolf pack in the next iteration is determined by the following formula:

Where X ₁ , X ₂ , and X ₃ are the α, β, and δ wolf positions, respectively;

Step S17: Perform Cauchy mutation operation on α wolf according to the following formula to determine whether the maximum number of iterations has been reached. If not, return to step S12 to continue the calculation;

_Xg (t)＝ _Xg (t)+η×C(0,1)

in

Where _Xg (t) is the global optimal solution in generation t; η is the mutation weight; λ is the adjustment factor; C(0,1) is the random number of the Cauchy mutation with a proportional parameter of t=1.

In step 2, the objective function of the ICGWO algorithm is:

The mean square error between the ELM model predicted output value and the measured value reflects the model prediction accuracy and generalization performance. Therefore, this paper selects the optimization objective function to minimize the ELM output root mean square error f _RMSE , which can effectively improve the fitting accuracy and generalization performance of the ELM regression model; the objective function is defined as follows:

Wherein, j=1,…,N, L is the number of hidden layer nodes, g(x) is the activation function, _wj =[ _wj1 , _wj2 ,…, _wjn ] ^T is the weight between the jth hidden layer node and the input node, _βj =[ _βj1 , _βj2 ,…, _βjm ^{] T} is the weight between the jth hidden layer node and the output node, _bj is the threshold of the jth hidden layer node; _xi =[ _xi1 , _xi2 ,…, _xin ] ^T∈Rn represents the input data, _yi =[ _yi1 , _yi2 ,…, _yim ^{] T∈Rm} represents the expected output value.

The specific steps of the present invention are as follows:

S1. Collect battery data including terminal voltage, current, internal resistance, open circuit voltage and SOC, obtain sample data under multiple different working conditions, and establish training data sets and test data sets.

S2. Construct a single-layer feedforward neural network consisting of an input layer, a hidden layer, and an output layer, and set the weights and thresholds of each neuron.

S3. Set the number of populations and problem dimensions in the gray wolf algorithm, and construct a gray wolf population representing the ELM model. Set the maximum number of iterations T _max of the gray wolf algorithm, the fitness function F, randomly initialize the positions of the population individuals in the gray wolf algorithm, and use the top three wolves with the best fitness values as the initial α, β, and δ wolves.

S4. Assign the weights and thresholds in the constructed ELM to the values of each dimension in the wolf pack position in order. Input the training set described in S1 into the ELM, and then calculate the error value of the predicted SOC through the fitness function F. If the number of iterations does not reach the maximum number of iterations T _max , then the number of iterations is t+1, otherwise the iteration is terminated and the result is output.

S5. Update the location of each wolf species in the gray wolf pack

S6. Perform Cauchy mutation on α wolf

S7. Take the output of the algorithm as the optimal individual of the wolf pack, assign the value of each dimension of the individual to the weight and threshold in ELM, and then you can get the optimized ELM model.

S8. Import the test set described in step S1 into the optimized IGGWO-ELM algorithm for training to obtain the power lithium battery SOC prediction result.

The steps in step S1 specifically include:

S9. First, use the battery tester and the host computer to test the current, voltage, temperature and other data of the lithium battery under various working conditions, and then calculate the SOC of the battery at each moment according to the ampere-hour integration method:

In the formula, z(t) represents the SOC of the battery at time t, i(t) represents the current of the battery at time t, _Cn represents the nominal capacity of the lithium battery, and _ηi represents the Coulomb coefficient of the lithium battery. The extreme learning machine model is constructed by taking current, voltage, and temperature as input and SOC as output.

The steps in step S2 specifically include:

设定ELM隐层节点数L，激励函数g(x)；Given a sample dataset

Set the number of ELM hidden layer nodes L and the activation function g(x);

其中x_i＝[x_i1,x_i2,…,x_in]^T∈Rⁿ表示输入数据，y_i＝[y_i1,y_i2,…,y_im]^T∈R^m表示为期望输出值。如果含有L个隐层节点的ELM模型能够学习N个训练样本，且无残差存在，则存在w_i使得：Furthermore, the training sample set

Where x _i = [ _xi1 , x _i2 , …, x _in ] ^T ∈ R ⁿ represents the input data, and y _i = [y _i1 , y _i2 , …, y _im ] ^T ∈ R ^m represents the expected output value. If the ELM model with L hidden nodes can learn N training samples and there is no residual, then there exists _wi such that:

Among them, j = 1,…,N, L is the number of hidden layer nodes, g(x) is the Sigmoid function, w _j = [w _j1 ,w _j2 ,…,w _jn ] ^T is the weight between the jth hidden layer node and the input node, β _j = [β _j1 ,β _j2 ,…,β _jm ] ^T represents the weight between the jth hidden layer node and the output node, and b _j is the threshold of the jth hidden layer node.

Furthermore, the formula can be rewritten as a matrix vector form: Hβ＝Y

in:

Therefore, the connection weight β between the hidden layer nodes and the output layer nodes can be obtained by the minimum two-norm least squares solution of the following formula:

Where H ⁺ is the generalized inverse matrix of the hidden layer output matrix H.

The step of randomly initializing the positions of individuals in the population in the grey wolf algorithm in step S3 specifically includes:

Using the Tent mapping equation to generate the initial population can make the initial individuals distributed as evenly as possible in the search area. The specific formula is:

The fitness function F in step S4 specifically refers to:

The objective function is optimized to minimize the root mean square error (f _RMSE ) of the ELM output, which can effectively improve the fitting accuracy and generalization performance of the ELM regression model. The objective function is defined as follows:

The updating of the positions of the various species of wolves in the gray wolf pack in step S4 specifically refers to:

The position update formula of each wolf species in the gray wolf pack is as follows:

D＝|C· _Xp (t)-X(t)|

X(t+1)＝ _Xp (t)-A·D

Where t is the current iteration number, D is the distance between the wolf and the prey, A and C are coefficient vectors, _Xp (t) is the position of the prey, and X is the position of the wolf. A and C are calculated by the following formula:

A＝2a·r ₁ -a

C＝2· _r2

Where a is a convergence factor that decreases linearly from 2 to 0, calculated by equation (5), and _r1 and _r2 are random vectors in [0,1].

Among them, t is the current iteration number, and T _max is the maximum iteration number.

The Cauchy mutation operation on the α wolf in step S5 specifically refers to:

The Cauchy mutation operator is used to maintain a balance between the diversity of the population and the convergence of the algorithm during the evolution process. The mutation formula is as follows:

_Xg (t)＝ _Xg (t)+η×C(0,1)

Where, X _g (t) – the global optimal solution in generation t;

η——variation weight;

λ——regulation factor;

C(0,1)——Cauchy mutation of random numbers with scale parameter t=1.

Specifically:

S1: In order to achieve the purpose of estimating the SOC of the power battery, the algorithm flow of the present invention is shown in FIG1 , the battery state of charge ICGWO-ELM estimation model flow is shown in FIG2 , and the operating condition data of the power battery adopts the American Urban Dynamic Drive Cycle (UDDS), and its time-speed diagram is shown in FIG3 .

S2: Run the simulation test conditions once to obtain the corresponding data, select current, terminal voltage, internal resistance, open circuit voltage and battery temperature as input, and SOC as output to form a 5-input single-output ELM model. The sampling frequency is set to 1Hz. There are 1370 sets of samples in the UDDS condition. Since the data of lithium batteries is inaccurate due to the polarization effect in the low SOC stage, the first 1000 sets of data are selected as samples. 750 sets of data are selected as training sets, and the remaining 250 sets of data are used as test sets.

Since the order of magnitude of each input and output differs greatly, the samples used for training are first normalized by the following formula before training.

Where X is the data to be normalized; X _min is the minimum value in the normalized data set; X _max is the maximum value in the normalized data set; X′ is the normalized data.

S3: To further verify the effectiveness of the ICGWO-ELM battery SOC estimation model proposed in this paper, the optimal initial input weights and hidden layer thresholds of the ELM regression model are first determined using the ICGWO algorithm based on the training samples, and then the battery SOC is predicted using the test data samples. Finally, the prediction results are compared with those of the IGWO-ELM, GWO-ELM, PSO-ELM, and ELM models. In the simulation experiment, the population size N = 30, the maximum number of iterations T _max = 200, the adjustment factor λ = 1000 in the ICGWO algorithm, and the learning factor c ₁ = c ₂ = 2 in the PSO algorithm. The number of hidden layer nodes in the four regression models of ICGWO-ELM, GWO-ELM, PSO-ELM, and ELM is L = 10, and the results are compared in Figures 4 and 5.

S4: In order to further clearly reflect the accuracy of the SOC estimation result, the relative error is calculated by the following formula for comparison, and the result is shown in Figure 6.

——第i项数据的预测值。In the formula, ε is the relative error, y _i is the true value of the i-th data,

——The predicted value of the ith data item.

As can be seen from the figure, the ELM model has the largest error throughout the process, and the accuracy of the PSO-ELM model is greatly improved compared with the ELM, but divergence occurs in the middle and late stages, which increases the error. The overall prediction effect of GWO-ELM, IGWO-ELM and ICGWO-ELM is better than that of the first two algorithms. Although the error of GWO-ELM in the middle and late stages of prediction is lower than that of PSO-ELM, it is still higher than that of IGWO-ELM and ICGWO-ELM. The ICGWO-ELM model successfully controls the relative error within 2%, and the predicted values at almost all sample points are very close to the actual values. This shows that the SOC estimation effect of the ICGWO-ELM model is significantly better than that of the other four models.

S5. In order to more accurately evaluate the performance of the proposed model, the error indicators mean absolute error (MAE), mean relative error (MRE) and root mean square error (RMSE) are used to evaluate the soft sensor model. The expressions are as follows:

表示第i个样本的预测值。性能指标计算结果如表1所示：Where N represents the number of samples; _yi represents the actual value of the i-th sample;

Represents the predicted value of the i-th sample. The performance index calculation results are shown in Table 1:

Table 1 Performance comparison results of four battery SOC estimation models

It can be seen from the table that ICGWO has the best performance and the highest accuracy, which shows that the ICGWO-ELM battery SOC estimation model has high prediction accuracy and strong generalization ability.

Claims (4)

1. An ICGWO optimization ELM-based lithium battery SOC estimation method is characterized by comprising the following steps of:

step 1: aiming at the problems that the early maturing phenomenon and local convergence are frequently encountered in the late evolution stage of the gray wolf algorithm due to rapid decline of population diversity, an improved gray wolf algorithm with good global optimum searching performance is constructed by introducing an improvement strategy;

step 2: and (3) optimizing hidden layer threshold values and input weight parameters of the limit learning machine by utilizing an improved wolf algorithm and minimizing the root mean square error of the model as a target function, and establishing an SOC estimation model of the lithium battery based on ICGWOO optimized ELM.

2. The ICGWO optimization ELM-based lithium battery SOC estimation method as claimed in claim 1, wherein the method comprises the following steps: in step 1, the improvement strategy of the ICGWO algorithm is as follows:

selecting a Tent mapping equation to generate an initial population by using the randomness and the ergodicity of the chaotic sequence;

secondly, determining the position of the wolf pack according to a nonlinear factor adjusting strategy so as to increase the local searching capacity of the algorithm; and finally, carrying out Cauchy variation disturbance on the current optimal solution, increasing the population diversity and improving the global search capability of the algorithm.

3. The ICGWO optimization ELM-based lithium battery SOC estimation method as claimed in claim 1, wherein the method comprises the following steps: in step 1, the specific steps of the ICGWO algorithm are as follows:

step S11: setting an initial population size N and a maximum iteration number T of an ICGWO algorithm _max Adjusting parameters such as a coefficient lambda and the like, and determining upper and lower limits of an optimization variable;

step S12: generating an initial wolf population satisfying variable upper and lower limit constraints by using a Tent chaotic mapping formula, and enabling t =1;

in the formula, x _t Is the position of the tth wolf;

step S13: calculating the fitness of each wolf, selecting the first three wolfs with the lowest fitness as alpha, beta and delta respectively, and recording the rest wolfs as omega wolfs;

step S14: judging whether the maximum iteration times are reached, if so, outputting the corresponding fitness of the alpha wolf, finishing the algorithm, otherwise, jumping to the step S15;

step S15: according to the formula

Calculating a nonlinear convergence factor a, then according to the formula a =2a · r ₁ -a and C =2 · r ₂ Calculating coefficient vectors A and C;

in the formula, a is a nonlinear convergence factor, and A and C are coefficient vectors;

step S16: the positions of the various species of wolfs of the gray wolf cluster are updated in turn according to the following formula:

D＝|C·X _p (t)-X(t)|

X(t+1)＝X _p (t)-A·D

wherein t is the current iteration number, D represents the distance between the wolf and the prey, A and C are coefficient vectors, and X _p (t) is the position of the prey, X represents the position of the wolf; in particular the starting point in the next iteration of the wolf pack is determined by the following formula;

in the formula, X ₁ ，X ₂ ，X ₃ Alpha, beta and delta wolf positions, respectively;

step S17: carrying out Cauchy mutation operation on the alpha wolf according to the following formula, judging whether the maximum iteration times is reached, and returning to the step S12 to continue operation if the maximum iteration times is not reached;

X _g (t)＝X _g (t)+η×C(0，1)

wherein

In the formula, X _g (t) is the global optimal solution at the t generation(ii) a Eta is the variation weight; lambda is an adjustment factor; c (0, 1) is a random number of cauchy variation with a scaling parameter t = 1.

4. The ICGWO-optimized ELM-based lithium battery SOC estimation method of claim 1, wherein: in step 2, the objective function of the ICGWO algorithm is as follows:

the mean square error between the predicted output value and the measured value of the ELM model reflects the prediction precision and generalization performance of the model, so that the optimized objective function is selected to output the mean square error f by the ELM _RMSE Minimization is realized, so that the fitting precision and the generalization performance of the ELM regression model can be effectively improved; the objective function is defined as

Wherein j =1, \8230, N and L are hidden node numbers, g (x) is a stimulus function, and w _j ＝[w _j1 ，w _j2 ，…，w _jn ] ^T Is a weight, β, between the jth hidden node and the input node _j ＝[β _j1 ，β _j2 ，…，β _jm ] ^T Representing the weight between the jth hidden node and the output node, b _j A threshold value of the jth hidden node; x is the number of _i ＝[x _i1 ，x _i2 ，…，x _in ] ^T ∈R ⁿ Representing input data, y _i ＝[y _i1 ，y _i2 ，…，y _im ] ^T ∈R ^m Expressed as the desired output value.

Images